Outliers or Poor Indicators of Climate

Chris Martz decided to make a point about extreme temperatures by posting a map with the hottest August temperatures ever recorded in each state of the US. Martz asks:

Did you know that 33 U.S. states set their “all-time” August high temperature records before 1955?  Of those, eight Midwestern states set their August records in 1936 alone while seven set theirs in 1918.

Here's the map he shared.

Chris Martz's Map on X

This I suppose would be interesting if we were considering when outliers appear to clump up. There's probably something meaningful in the fact that 8 states had their warmest August record temperature in 1936 and 7 had their warmest August record in 1918. We all know about the Dust Bowl, during which human activity was responsible for causing extremely high daily maximum temperatures (Tmax) during the summer months, especially in the Midwest. GHG-induced warming causes Tmax to increase but it causes Tmin to warm even more rapidly. Human activity made the Dust Bowl years much hotter than they otherwise would be, and AGW is making heat waves much worse as well. But we need to consider how meaningful this map is in discussions about climate change.

A few things should be obvious about maps like this:

1. There are only 51 data points in this graph. GHCNd contains about 10,000 stations. The number of stations is not consistent across CONUS history since 1895, but nClimDiv operates on a grid with 344 cells, essentially a 5 km resolution grid. So to simplify my point, let's work with the 344 climate divisions. The month of August has 31 days, and there have been 131 Augusts since 1895. Each climate division records 2 two temperatures daily, a Tmax and a Tmin. So that means there have been 31*131*344*2 = 2,793,968 temperature data points for CONUS since 1895 (more if we include Alaska and Hawaii). So this map depends on 0.0018% of the total number of data points recorded for CONUS.
2. All 51 of these datapoints are outliers. They are not indicative of what was normal or typical when the temperature was recorded, and we learn nothing about how frequent temperatures are slightly below the record temperature. The saying that "climate is what you expect; weather is what you get" is relevant here. The fact that freakishly hot temperatures happened in 1936 is not an argument against the fact that temperatures that used to be considered extreme are now becoming more common.
3. Once a record high is set, it's more difficult to break it. This means that we should expect a decrease in the frequency of records being broken with time, especially if stationarity can be assumed. The fact that an extreme outlier in of 120°F occurred in Texas in August 1936 means that any new record will have to exceed 120°F. If Tmax readings of 115-119°F are becoming far more common in TX, you wouldn't be able to tell by looking at this map. All that information his hidden by Martz's choice to show only one outlier per state.

There are better ways to look at the data that don't require eliminating 99.988% of the available data. Here are a couple ways that can help us see if extreme heat is becoming more common and expected.

Andrew Dessler took all the gridded CONUS Tmax temperature data in Berkeley Earth and plotted the distribution of the top 5% of temperatures. The plot below shows when the top 5% of gridbox-days occurs. Clearly the 1930s (and 1936) shows a large spike in gridbox days in the top 5% of recorded Tmax. However, over the last 30 years or so, multiple years have a higher share of the gridbox-days. In other words, The top 5% of Tmax readings occur more frequently in recent decades than in the dust bowl years. 

Zeke Hausfather plotted the frequency of Tmax reading above 95°F since 1895 in the CONUS daily temperatures within Berkeley Earth. He then shows the 11-year mean for CONUS and finds that while there was a bump in these readings during the 1930s, they are even more frequent in the 21st century.

And of course this is just for the US. Since the Dust Bowl was basically localized to the US, and since the US is only about 2% of the globe, we can check to if the extremes seen in the 1930s have much an effect on the globe. Here's a map of anomalies for the last 10 years against a 1930s baseline. The blue portion of the globe is the part that was warmer in the 1930s, and it's almost entirely localized to the US.

From ClimateBrink

It's unsurprising, therefore, that it turns out that days above 95°F, 100°F, and 105°F are becoming much more frequent globally, especially over the last 30 years or so.

And of course, none of this is unexpected given the fact that CONUS is warming, and Tmax has been increasing. Here is NOAA's data for CONUS annual Tmax.

August Tmax for CONUS is also warming, and despite freakishly hot Augusts in 1936 and 1937, both are eclipsed by 2011. The point here though, is not that 2011 beat 1936. The point is that hot temperatures in August are becoming more common in CONUS.

Chris Martz's map is essentially meaningless when it comes to assessing how AGW is affecting extreme temperatures, both in CONUS and across the globe. It may look convincing to his followers, but outliers are indicative of freak events, and freak events happen with or without AGW. What we observe with climate change is that what used to be extreme temperatures are now becoming normalized, and what is extreme for us now was freakishly uncommon in the past.